User Guide On Homography Refinement
Image Stitching for NVIDIA®Jetson™ |
---|
Before Starting |
Image Stitching Basics |
Overview |
Getting Started |
User Guide |
Resources |
Examples |
Spherical Video |
Performance |
Contact Us |
Homography Refinement
The stitcher has the capability of reestimate the homographies on the fly. It means that the stitcher can adjust the images based on the current environment information.
Properties
async-homography-enable
:- if set to true, the homography refinement feature is enabled.
min-matches
:- is the minimum number of matches that must be present in the images to compute a new homography.
(Min: 4)
- is the minimum number of matches that must be present in the images to compute a new homography.
match-ratio
:- is the ratio to match the points between the images.
(Min: 0.0, Max: 1.0)
- is the ratio to match the points between the images.
homography-queue-size
:- the number of the accumulated homographies from where an average homography is computed and used in the image stitching.
(Min: 1)
- the number of the accumulated homographies from where an average homography is computed and used in the image stitching.
kernel-size
:- the size of the Gaussian kernel to filter the noise from the images.
(Min: <1,1>)
- the size of the Gaussian kernel to filter the noise from the images.
kernel-sigmas
:- the sigmas of the Gaussian kernel. Defaults to
<0,0>
which is the recommended as OpenCV will compute the sigmas based on the kernel size.
- the sigmas of the Gaussian kernel. Defaults to
How to adjust the properties
The following table shows a guide on how to vary the refinement parameters in order to improve the quality of the homographies, or to get refinements more often. The best values will depend on your setup (cameras and their overlap) and the environment (light conditions and features in the scene).
Property | Better homographies | Faster refinement |
---|---|---|
min-matches | ↑ | ↓ |
match-ratio | ↓ | ↑ |
homography-queue-size | ↑ | ↓ |
Also, if your images present noise, you can increase the kernel-size
to filter the images and then compute better homographies.